Systematic Review for Water Network Failure Models and Cases

University of Arkansas, FayettevilleScholarWorks@UARK

Theses and Dissertations

12-2017

Systematic Review for Water Network FailureModels and CasesYufei GaoUniversity of Arkansas, Fayetteville

Follow this and additional works at: http://scholarworks.uark.edu/etd

Part of the Hydraulic Engineering Commons, Industrial Engineering Commons, and theStructural Engineering Commons

This Thesis is brought to you for free and open access by ScholarWorks@UARK. It has been accepted for inclusion in Theses and Dissertations by anauthorized administrator of ScholarWorks@UARK. For more information, please contact [email protected], [email protected].

Recommended CitationGao, Yufei, "Systematic Review for Water Network Failure Models and Cases" (2017). Theses and Dissertations. 2608.http://scholarworks.uark.edu/etd/2608

http://scholarworks.uark.edu?utm_source=scholarworks.uark.edu%2Fetd%2F2608&utm_medium=PDF&utm_campaign=PDFCoverPages

http://scholarworks.uark.edu/etd?utm_source=scholarworks.uark.edu%2Fetd%2F2608&utm_medium=PDF&utm_campaign=PDFCoverPages

http://scholarworks.uark.edu/etd?utm_source=scholarworks.uark.edu%2Fetd%2F2608&utm_medium=PDF&utm_campaign=PDFCoverPages

http://network.bepress.com/hgg/discipline/1087?utm_source=scholarworks.uark.edu%2Fetd%2F2608&utm_medium=PDF&utm_campaign=PDFCoverPages



http://scholarworks.uark.edu/etd/2608?utm_source=scholarworks.uark.edu%2Fetd%2F2608&utm_medium=PDF&utm_campaign=PDFCoverPages

mailto:[email protected],%[email protected]

Systematic Review for Water Network Failure Models and Cases

A thesis submitted in partial fulfillment

of the requirements for the degree of

Master of Science in Industrial Engineering

by

Yufei Gao

Wuhan University of Technology

Bachelor of Science in Material, 2014

December 2017

University of Arkansas

This thesis is approved for recommendation to the Graduate Council.

________________________________

Shengfan Zhang, Ph.D.

Thesis Director

________________________________

Haitao Liao, Ph.D.

Committee Member

________________________________ ________________________________

Manuel D. Rossetti, Ph.D. Wen Zhang, Ph.D.

Committee Member Committee Member

Abstract

As estimated in the American Society of Civil Engineers 2017 report, in the United States, there

are approximately 240,000 water main pipe breaks each year. To help estimate pipe breaks and

maintenance frequency, a number of physically-based and statistically-based water main failure

prediction models have been developed in the last 30 years. Precious review papers focused more

on the evolution of failure models rather than modeling results. However the modeling results of

different models applied in case studies are worth reviewing as well.

In this review, we focus on research papers after Year 2008 and collect latest cases without

repetition. A total of 64 papers are qualified following the selection criteria. Detailed information

on models and cases are summarized and compared. Chapter 2 provides a summary and review of

failure models and discusses the limitation of current models. Chapter 3 provides a comprehensive

review of collected cases, which include network characteristics and factors. Chapter 4 focuses on

the main findings from collected papers. We conclude with insights and suggestions for future

model selection for pipe failure analysis.

©2017 by Yufei Gao

All Rights Reserved

Contents

Chapter 1 Introduction ........................................................................................................ 1

1.1 Motivation ................................................................................................................. 2

1.2 Objectives and Design of Systematic Review........................................................... 3

1.3 Organization of the Thesis ........................................................................................ 5

Chapter 2 Review of Water Network Failure Models ........................................................ 6

2.1 Deterministic Models ................................................................................................ 6

2.2 Probabilistic Models .................................................................................................. 6

2.3 Stochastic model ..................................................................................................... 12

2.4 Artificial Intelligence and Machine Learning Methods .......................................... 13

2.5 Summary and Limitation of Previous Studies......................................................... 15

Chapter 3 Review of Contributing Factors ....................................................................... 17

3.1 Classification of Factors .......................................................................................... 17

3.2 Effect of Factors in Previous Papers ....................................................................... 19

3.3 Summary of Factors Considered in Different Failure Models ................................ 26

3.4 Factor Distribution Analysis .................................................................................. 34

Chapter 4 Summary of Key Findings in Previous Studies .............................................. 36

4.1 Factor Effect in Regression Models ........................................................................ 36

4.2 Model Results Review ............................................................................................ 38

Chapter 5 Summary and Conclusion ................................................................................ 40

5.1 Conclusions ............................................................................................................. 40

5.2 Limitation and Future Work .................................................................................... 42

References ......................................................................................................................... 43

Appendix I ........................................................................................................................ 47

Appendix II ....................................................................................................................... 53

1

Chapter 1 Introduction

Drinking water is delivered via one million miles of pipes across the U.S. Aging pipe has been one

of the major challenges facing the water industry due to the limitation of funding availability.

Many of those pipes were laid in the early to mid 20th century with a lifespan of 75 to 100 years

(American Water Works Association (AWWA), 2017). Using these average life estimates and

counting the years since the original installations shows that these water utilities will face

significant needs for pipe replacement over the next few decades. Some components in water and

sanitation conveyance systems in the United States and Europe are more than 100 years old

(AWWA 2017). Aging pipes present many technical limitations for effective water provisioning.

Firstly, degradation of infrastructure system integrity leads to system losses and water leaks. The

water lost in the conveyance process is often referred to as “nonrevenue water” because it leaves

the system prior to the water meter, which is generally used to define cost paid by the user.

Secondly, supplied water by pipes with breaks generally carries a higher risk of contamination,

which could lead to various potential health impacts for users. As estimated in the American

Society of Civil Engineers (ASCE) 2017 Infrastructure Report Card, in the United States, there are

approximately 240,000 water main pipe breaks each year (ASCE 2017). As a result, 10% to 30%

of total water is non-revenue water, while in England this value has recently been estimated to be

25% (ASCE 2015). It is projected that above 1 million miles of water mains need replacement, as

estimate by AWWA (2017). The replacement cost is estimated to be approximately $1 trillion to

maintain and expand service to meet demand over the next 25 years (ASCE 2017). However,

2

constrained by the limited resources available, efficient maintenance and management of water

infrastructure, particularly pipe maintenance and repair in the distribution system, is challenging

but imperative.

To deal with this problem, a number of physically-based and statistically-based water main failure

prediction models have been developed in the last 30 years. Physical models predict breaks by

simulating the mechanics of pipe failure and the capacity of a pipe to resist failure. Statistical

models are developed with historical data on pipe breaks to identify failure patterns, and they

extrapolate these patterns to predict future pipe breaks (MJ Nishiyama, 2013).

1.1 Motivation

Most papers about water network failure focused on failure model development and validation,

with case studies using one or more real database of networks. Previous review papers on failure

models summarized the evolution of models, compared the differences between various models,

and defined a variety of classification of models (Clair & Sinha, 2012; Nishiyama & Filion, 2013).

However, all these discussion and comparisons did not mention much about the applied cases. The

application of each single case and the specific conclusion for real data are seldom reviewed in the

past 20 years. The characteristics of cases covers a lot of information such as region (water and air

temperature, the depth of pipes), pipeline scale (the number of pipes ranging from tens to

thousands), date of construction (which is highly associated with pipe material used), the state of

maintenance and record (the frequency of maintenance and the integrity of maintenance record).

3

A better understanding of the relationship between failure prediction and network characteristics

would be useful for failure model selection when an analyst works on another similar real case. In

addition, previous conclusions from case studies could be used as validation for prediction and

direction for analysis in the future. Therefore, a comprehensive review for water network cases

and results is necessary and worthwhile.

1.2 Objectives and Design of Systematic Review

The overall goal of this paper is to provide a comprehensive review of recent water network failure

models and cases. In this review, we focus on research papers after Year 2008 and collect latest

cases without repetition. Detailed information on models and cases such as attributes of networks

considered in the models are summarized and compared. Papers selected in this research are

searched by key words: pipe failure, water distribution, failure prediction, pipe break, pipe

deterioration. A few papers were collected from the citation of pervious review papers (Genevieve

Pelletier 2003; Berardi et al. 2008). After paper collection, case screening was processed by several

principles: remove papers before 2008 and keep papers that have the case study part. According

to the flow diagram in Figure 1, 64 cases were collected in total.

4

Figure 1. Selection Criteria Flow Diagram

5

1.3 Organization of the Thesis

The remainder of the thesis is organized as follows. Chapter 2 provides a brief review of failure

models in previous review papers and discusses the limitation of current models. Chapter 3

provides a comprehensive review of collected cases which include network characteristics and

factors. Chapter 4 focuses on the main findings from collected papers. Common points are

extracted as insights and suggestion for future model selection in pipe failure.

6

Chapter 2 Review of Water Network Failure Models

During the last three decades, researchers developed different models to predict the failure of water

pipes for a reliable infrastructure management. These failure prediction models can be classified

into four categories: deterministic, statistical, stochastic, artificial intelligence models. In the next

few sections, we first review each category in detail with a focus on the studies in the last decade

and then summarize in Section 2.5.

2.1 Deterministic Models

Deterministic models usually are used in cases where the relationship between inputs and output

is clear. In two approaches the deterministic models can be applied: empirical and mechanistic.

Empirical approach tries to find the relations between failure rates as the output and the features

and attributes of a group of pipes as the inputs, while the mechanistic approach can forecast the

remaining useful life of an individual asset (just one pipe). Many papers (Kwietnieswki et al. 1993;

Kowalski 2013; Kutylowska 2014) used a similar definition of failure rate. The value of λ is

determined from operational data using number of pipe failures in unit time interval divide average

pipeline length in a time period and the observation time. The problem of these models is that a

deterministic model can be applied just in a specific location (Clair and Sinha 2012).

2.2 Probabilistic Models

Probabilistic models analyze the probability of an event occurring (Creighton 1994). The

7

probability of occurrence is one and the probability of the event that cannot happen is zero. The

other probability of occurrence should be between 0 and 1 (Mitrani 1998). Information about asset

conditions and attributes are required to develop a probabilistic model. The output or dependent

variable would be a range of values instead of the specific number. These models need extensive

data and typically used in infrastructure assets (Clair and Sinha 2012). It should be noted that the

probabilistic approach commonly increases the computational complexity of the models (Moglia

2007).

The Evolutionary Polynomial Regression (EPR) technique was first presented by Giustolisi and

Savic (2006). The technique utilizes the huge potential of conventional numerical regression

techniques and the strength of Genetic Algorithm in solving optimization problems (Xu et al. 2011).

Later, this approach was used by other researchers in several engineering fields. Savic et al. (2006)

and Ugarelli et al. (2008) used EPR to model the sewer pipe failures. Berardi et al. (2008) and Xu

et al. (2011) applied the EPR to develop deterioration models for water distribution networks.

Rezania et al. (2008) utilized the EPR methodology to evaluate the uplift capacity of suction

caissons and shear strength of reinforced concrete deep beams. Elshorbagy and El-Baroudy (2009)

compared the EPR and Genetic Programming to develop the prediction model of soil moisture

response.

Guistolisi and Savic (2009) tested the EPR-MOGA (an improved EPR) to develop a model to

forecast the groundwater level based on the amount of rainfall each month. El-Baroudy et al. (2010)

utilized the EPR to develop the evapotranspiration process then compared the efficiency of

8

Evolutionary Polynomial Regression to Artificial Neural Networks (ANNs) and Genetic

Programming (GP). Markus et al. (2010) applied EPR, ANNs and the naive Bayes model to

forecast weekly nitrate-N concentrations at a gauging station. Ahangar-Asr et al. (2011) applied

EPR to predict mechanical properties of rubber concrete. Fiore et al. (2012) used EPR to provide

the predicting torsional strength model of reinforced concrete beams.

Moglia et al. (2007) developed a physical probabilistic failure prediction model based on the

fracture mechanics of cast 30 iron water pipes. The random independent variables were added to

the inputs, and then Monte-Carlo simulation technique was applied to deal with the computational

complexity of the model. The developed model without failure data, degradation and load data,

was not capable of estimating failure rates of water pipes. Whereas, with these data, it can predict

failure rates more accurately.

Li et al. (2009) used the mechanically-based probabilistic model to predict remaining useful life

and failure probability of buried pipes. They considered the effect of random inputs and used

Monte-Carlo simulation framework to calculate cumulative distribution function (CDF) of

remaining useful life of pipelines. But, they did not consider the correlation of defects for a pipeline

having more than one corrosion defects. Also, they found CDF more suitable than probability

density function (PDF) and reliability index in describing the probability of failure.

It should be mentioned that this technique requires a large historical dataset that contains a number

of data points collected over a period to develop a promising statistical model (Clair and Sinha

9

2012). There has been an extensive effort during the past decades to develop the failure rate

prediction model by using statistical approach.

Berardi et al. (2008) developed a water pipe deterioration model using Evolutionary Polynomial

Regression. As it is mentioned before, they used a dataset that was classified into homogeneous

groups based on the age and diameter of the pipe. The developed model can predict the number of

breaks in each group. Then, for predicting the failure rate for each pipe, a general structural

deterioration model based on EPR aggregated model was developed.

Wang et al. (2009) utilized five multiple regression models for a range of pipe materials (gray cast

iron, ductile iron without lining, ductile iron with lining, PVC, and hypericin) to forecast the annual

failing rate of individual water pipe rather than a homogeneous group. The overall model

robustness was measured by F-test and the significant of each independent variable was measured

by t-test. The model was validated using 20% of their collected dataset that was randomly selected.

Wang et al. (2010) employed the Bayesian inference to assess the condition of water pipes. Ten

factors from three pipe materials (cast iron, ductile cast iron, and steel) were used to generate factor

weight. According to the results of these experiments, the age of pipe is the most critical variable

28 while, the model was not sensitive to some factors like trench depth, electrical recharge, and

some road lanes.

Xu et al. (2011) developed two prediction models for failure rate using Evolutionary Polynomial

Regression and Genetic Programming, and then they compared the results of these two models.

10

Results were measured based on; 1) error between predicted and actual data, 2) parsimony of

generated equation, and 3) ability to justify the generated equations based on the engineering

knowledge. The results showed that EPR has some advantages over GP in equation uniformity and

parameters estimation, while GP was better to find the complex relations.

Osman and Bainbridge (2011) employed two statistical deterioration models to predict future

failures of water pipes: rate-of-failure models (ROF) and transition-state (TS) models. ROF model

extrapolates the failure rate for a specific group of water pipes that were classified based on age

and some environmental factors. This model does not differentiate the times between successive

pipe breaks for an individual segment while, the transition-state model focuses on finding the time

between successive failures for the water pipes. TS models are dependent on the availability of

sufficient and accurate data, but ROF models can be applied to limited historical data. The stresses

in the buried pipes, which increase the probability of pipe failure, might be caused by the ground

movement.

Kabir et al. (2015) presented Bayesian Model Averaging (BMA) method to select the most critical

explanatory variables. Then the Bayesian Weibull Proportional Hazard 29 Model (BWPHM) is

applied to provide the survival curves and to forecast the failure rate of two pipe types: cast iron

and ductile iron.

Kabir et al. (2014) assessed the risk of failure of metallic water pipes using a Bayesian Belief

Network (BBN). Bayesian Belief Network can be interpreted as a probabilistic graphical model

11

that can represent a collection of some covariates and their probabilistic relationships. This model

recognizes the most vulnerable and sensitive pipe segments through the water pipe networks. The

proposed model is good just for small to medium utilities with limited data.

Jenkins et al. (2014) tried to address the problem of limited, incomplete, or uncertain data in water

distribution networks. Two main modification were added to Weibull hazard rate models (WPHM)

to improve the prediction performance of the models: the expert opinion and the spatial analysis.

But these two modifications were not tested in the other utilities.

Francis et al. (2014) analyzed the water distribution systems to develop a pipe breaks prediction

model using Bayesian Belief Networks (BBNs). They illustrated that assessing water pipe network

is not only important for the failure prediction model but also is crucial for avoiding water loss and

water quality degradation.

Kabir et al. (2015) stated that uncertainty regarding quality and quantity of databases became a

major concern for failure prediction model development of infrastructure assets. Thus, they tried

to reduce these uncertainties by developing failure prediction model for water mains using a new

Bayesian belief network based data fusion model. The proposed model can identify the most

vulnerable and sensitive pipe in the entire network, as well as the total number of pipes that require

the immediate and appropriate action like maintenance, rehabilitation, and replacement

Konstantinos Kakoudakis et al. (2017) presented a new approach for improving pipeline failure

predictions by combining a data-driven statistical model, i.e. evolutionary polynomial regression

12

(EPR), with K-means clustering. The EPR is used for prediction of pipe failures based on length,

diameter and age of pipes as explanatory factors. Individual pipes are aggregated using their

attributes of age, diameter and soil type to create homogenous groups of pipes. The created groups

were divided into training and test datasets using the cross-validation technique for calibration and

validation purposes respectively. The K-means clustering is employed to partition the training data

into a number of clusters for individual EPR models

2.3 Stochastic model

A stochastic model is a tool for estimating probability distributions of potential outcomes by

allowing for random variation in one or more inputs over time. Poisson process, nonhomogeneous

Poisson process, Yule process are classified in this type. To see occurrences of pipe breaks over a

certain period as stochastic point processes is one of the common ways to model them. (Kleiner

and Rajani, 2001; Gat and Eisenbeis, 2001). One of the point processes that is often used is the

non-homogeneous Poisson process (NHPP). This is because its great flexibility allows it to capture

the non-linear relationship of the break rate with time without giving up on the inclusion of suitable

pipe factors (Loganathan et al., 2002). Li Chik et al. (2016) used the NHPP, hierarchical beta

process (HBP), and a newly-developed Bayesian simple model (BSM) for short-term failure

forecasting with a few water utility failure data sets. After close analysis of the prediction curves,

they found that the performance of the three models are of great similarity in terms of pipe ranking.

However, compared with the other models, the BSM is relatively simpler, which has given it more

edges. The covariate, the number of known past breaks, can be very important when it comes to

13

the relative ranking of the pipes in the network. The NHPP and HBP are recommended if the total

number of failures in the network is required.

2.4 Artificial Intelligence and Machine Learning Methods

Artificial intelligence and machine learning models, which include Artificial Neural Networks

(ANN), Least square support vector machine method (LS-SVM) and Fuzzy set theory models,

become more and more popular in recent years due to its capability of dealing with complex data.

ANN is a method that can predict pipe failure and deterioration of infrastructure specially buried

pipes. The ANN follows the pattern of the human brain using its generalization capabilities. Thus,

this technique is able to process information even under large, complex, and uncertain environment.

The high-quality database is needed for supervised training and forecasting the future condition of

the pipes. Moreover, ANN needs several controlling factors including: number of hidden layers,

the number of neurons in each hidden layer, activation functions, the number of training epochs,

learning rate, and momentum term. However, ANN is considered as a “Black- 32 Box” technique.

Therefore, it is not able to provide insight into the relationship between dependent and

independents variables (Clair and Sinha 2012; Moselhi and Hegazy 1993, Atef et al. 2015, Shirzad

et al. 2014). Fuzzy Logic is a mathematical method in the field of artificial intelligence that widely

used by researchers to assign a value to a certain degree of membership instead of crisp values

such as zero and one. This method is known to deal with systems that are subject to uncertainties

and ambiguities. Fuzzy Logic is applicable in infrastructure assets like oil and gas, water, bridges

and highways (Siler and Buckley 2005, Clair and Sinha 2012).

14

Jafar et al. (2010) employed ANN to analyze the urban water mains. Six ANN models that predict

the failure rate of water pipes of a city in France were developed then, they tried to estimate the

optimal rehabilitation/replacement time for the same network. These prediction models were tested

and validated using cross validation. In the first part of this article, data collection was explained

then development and validation of ANN models were discussed. In the data collection part,

correlation and chi2 method were applied to select the most critical inputs.

Asnaashari et al. (2013) studied two different methods to forecast the water pipe’s failure rate.

Multi Linear Regression (MLR) and ANN were utilized, and their results were compared. The

value of R-Squared showed that the ANN model (R2=0.94) is more promising while the MLR

technique (R2=0.75) is just good enough for preliminary assessment. Shirzad et al. (2014)

compared the predictive performance of ANN and Support Vector Regression (SVR) in

forecasting the water pipe’s breakage rate. In addition, they investigated the effect of hydraulic

pressure (average and maximum hydraulic pressure values) on precision of predicting the pipe’s

failure rate. The results showed that the ANN model is more accurate, but it is not suitable for

generalization purposes. Thus, for management purposes, SVR might be more appropriate.

Kutyłowska (2014) predicted the failure rate of pipes in an urban water utility using ANN. They

employed quasi-Newton approach to train the model. The house connections and distribution pipes

are considered as two different sections in database, and the results for both were acceptable.

Aydogdu and Firat (2014) incorporated two methods: fuzzy clustering and Least Squares Support

15

Vector Machine (LS-SVM) in order to estimate the failure rate of water pipes. At first, they

developed failure rate estimation model using LS-SVM, and then fuzzy clustering method is

utilized to define nine sub-regions for predictive performance improvement of the model. For

model evaluation they employed some measurement indexes such as Correlation Coefficient (R),

Efficieny (E) and Root Mean Square Error (RMSE).

2.5 Summary and Limitation of Previous Studies

Table 1. Classification of Models and Corresponding Number of Cases in the Literature.

Classification Models

Number of

Cases

Deterministic Models Failure Rate 8

Probabilistic Models

Linear Regression, Evolutionary

Polynomial Regression (EPR),

Weibull Proportional Hazard Model

(WPHM), Bayesian Belief Network,

Weibull/Exponential Distribution (WE)

31

Stochastic Models Poisson process, NHPP, Yule process 12

Artificial

Intelligence/Machine

Learning

Artificial Neural Network(ANN), Fuzzy

Clustering; Least square support vector

machine method (LS-SVM)

15

16

As shown in Table 1, it is obvious that statistical models have been the most popular method for

failure prediction compared to other model types. Statistical models are used most frequently in

the latest 10 years although it mostly requires large number of available factors in dataset.

Deterministic model mainly refers to failure rate model which is usually only need the failure

number in a period of time, so it is easy to use than others. Stochastic models mostly used for data

that only include process information even under large, complex, and uncertain environment. In

most cases, datasets were clustered into different groups, based on the pipe material, and then one

model was developed for each group. Thus, there are several models just for one network that

might be tough to implement in the real world. Several techniques were utilized by the other

authors. Particularly, ANNs are commonly used in many studies. ANN is able to develop accurate

prediction models in complex and uncertain environments. However, EPR is selected because it

does not require large datasets for training and unlike ANN, it enables the recognition of

correlations among dependent and independent variables. Being as such, EPR is not a “Black-Box”

technique, but it is classified as a “Grey-Box” technique that can provide insight into the

relationship between inputs and the output. The process of development and selection of EPR

contains the engineering 36 knowledge that allows the user to understand the generated equations

and correlation between variables involved. In ANN, each attempt delivers particular output,

which can be different in other attempts with the same inputs and features, while, in EPR or

generally regressions, all similar attempts lead to the same equations as the output. Advantage

summary form

17

Chapter 3 Review of Contributing Factors

In this section, we summarize factors contributing to water network failure with two parts. We first

discuss classification of various factors in the literature, and then summarize the effects of

commonly used factors on network failures.

3.1 Classification of Factors

InfraGuide (2003) classified the factors contributing to the deterioration of water pipes to three

main categories: physical, environmental and operational. According to this classification,

physical factors include pipe material, pipe wall thickness, pipe age, pipe vintage, pipe diameter,

type of joints, thrust restraint, pipe lining and coating, dissimilar metals, pipe installation and pipe

manufacture. Pipe bedding, trench backfill, soil type, groundwater, climate, pipe location,

disturbances, stray electrical currents, and seismic activity are considered as the environmental

factors, while other researchers included rainfall, traffic and loading, and trench backfill as the

environmental factors as well (Kabir et al. 2015). The internal water pressure, transient pressure,

leakage, water quality, flow velocity, backflow potential, and O&M practices are examples of

operational factors. Others considered the nature and date of last failure (e.g., type, cause, severity),

nature of maintenance operations (e.g., TV inspections, pipe cleaning, cathodic protection), nature

and date of last repair (e.g., type, length), water quality and construction method as operational

factors that affect the failure rate of water pipes (InfraGuide 2003). The specific explanation of

each factor is shown in Table 2.

18

Table 2. Factors that contribute to water system deterioration (InfraGuide 2003)

Jon Røstum (2000) proposed another classification method which considered all the factors into 4

types: structural, external, internal, and maintenance. Table 3 provides more details about it.

Classification Factor Explanation

Physical

Pipe material Pipes made from different materials fail in different ways.

Pipe wall thickness Corrosion will penetrate thinner walled pipe more quickly.

Pipe age Effects of pipe degradation become more apparent over time.

Pipe vintage Pipes made at a particular time and place may be more vulnerable

to failure.

Pipe diameter Small diameter pipes are more susceptible to beam failure.

Type of joints Some types of joints have experienced premature failure (e.g.,

leadite)

Thrust restraint Inadequate restraint can increase longitudinal stresses.

Pipe lining and

coating Lined and coated pipes are less susceptible to corrosion.

Dissimilar metals Dissimilar metals are susceptible to galvanic corrosion.

Pipe installation Poor installation practices can damage pipes, making them

vulnerable to failure.

Pipe manufacture

Defects in pipe walls produced by manufacturing errors can make

pipes vulnerable to failure. This problem is most common in older

pit cast pipes.

Environmental

Pipe bedding Improper bedding may result in premature pipe failure.

Trench backfill Some backfill materials are corrosive or frost susceptible.

Soil type

Some soils are corrosive; some soils experience significant

volume changes in response to moisture changes, resulting in

changes to pipe loading. Presence of hydrocarbons and solvents

in soil may result in some pipe deterioration.

Groundwater Some groundwater is aggressive toward certain pipe materials.

Climate Climate influences frost penetration and soil moisture. Permafrost

must be considered in the north.

Pipe location Migration of road salt into soil can increase the rate of corrosion.

Disturbances

Underground disturbances in the immediate vicinity of an

existing pipe can lead to actual damage or changes in the support

and loading structure on the pipe.

Stray electrical

currents Stray currents cause electrolytic corrosion.

Seismic activity Seismic activity can increase stresses on pipe and cause pressure

surges.

Operational

Internal water

pressure, transient

pressure

Changes to internal water pressure will change stresses acting on

the pipe.

Leakage Leakage erodes pipe bedding and increases soil moisture in the

pipe zone.

Water quality Some water is aggressive, promoting corrosion

Flow velocity Rate of internal corrosion is greater in unlined dead-ended mains.

Backflow potential Cross connections with systems that do not contain potable water

can contaminate water distribution system.

O&M practices Poor practices can compromise structural integrity and water

quality.

19

Table 3. Factors affecting structural deterioration of water distribution pipes (Jon Røstum,

2000)

3.2 Effect of Factors in Previous Papers

In this section, we list and describe factors that are commonly identified to have the greatest impact

on pipe failure. Conclusions on these factors are also summarized.

Age and installation period

We can see the features of different failures in different phases of the installation process. After

the installation has been done, compared with time, these features will become more reliant on the

construction practice in each phase. The break rate in one construction phase might be higher than

that in another phase (Mosevoll, 1994). Sometimes, compared with pipes that are relatively young,

Structural

Variables

External/Environmental

Variables

Internal

Variables

Maintenance

Variables

Location Soil type Water

velocity Date of failure

Diameter Loading Water

pressure Date of repair

Length Groundwater Water

quality Location of failure

Year of

construction Direct stray current

Water

hammer Type of failure

Pipe material Bedding condition Internal

corrosion

Previous failure

history

Joint method Leakage rate

Internal

protection Salt for de-icing of road

External

protection Temperature

Pressure class External corrosion

Wall thickness

Laying depth

20

pipes that are older will be less prone to the effect of failures. For example, the walls of grey cast

iron pipes are produced by newer casting methods, and for the same external loads, these thinner

walls may cause more corrosion as well as more stress. It is only in the 1930s that we managed to

use backfill to extend the lifetime of pipes. Time has witnessed the improvements of the jointing

techniques, which make a higher degree of deflections at joints become possible. From 1950s to

1960s, when the number of houses just kept rising at a rapid rate, compared with the quality of the

buildings, people often placed more emphasis on the quantity. During this time, houses of a rather

bad quality as well as the poor skill of the construction workers could often be seen in the reports

(Sundahl, 1997). According to the report written by Andreou et al. (1987), compared with pipes

that failed at a later stage, pipes that failed in the initial stage usually have better performance.

Besides, Wengström (1993) has discovered that we cannot rely on pipe records to find out the age

dependency. This is also why he drew up the conclusion that it is possible for us to hide the age

dependency via repairs. In other words, after being repaired for around four times, pipes will

usually need to be taken out of the ground. sessing pipe

Corrosion

One of the causes of the need to replace a pipeline is corrosion as it can lead to degradation of

pipes that are made of grey cast iron, ductile iron and steel (Mosevoll, 1994). The internal corrosion

has great reliance on the features of the transported water (e.g. pH, alkalinity, bacteria and oxygen

content) while the external corrosion is reliant on the surroundings of the pipe (e.g. soil

characteristics, soil moisture, and aeration). However, Kumar Dey (2003) put forward the idea that

21

when we are doing the prediction, we also need to take into consideration the external corrosion

as its intensity will change according to the different conditions. In this regard, it is different from

the internal corrosion.

Diameter

The idea that pipes with small diameters are most prone to failures can be found in a large number

of literature works in the field. (Rajeev, 2003). Pipes with diameters that do not exceed or are equal

to 200mm failure the most often. The strength of smaller pipes is usually are usually smaller, and

their walls are also thinner. Also, they are usually constructed in a different way and their joints

are usually not as reliable. These are the reasons why smaller pipe dimensions fail more frequently

(Wengström, 1993). Another possible cause for this is the lower velocities in smaller pipes, which

can cause the suspended materials in the water to settle, and this can make it easier for the bacteria

to grow. (National Research Council. (2006)).

Pipe length

The length of pipes, regardless of which network they are in, varies from one to another. For long

pipes (e.g. >1000m), external conditions including the condition of the soil as well as the traffic

might be different depending on the pipe. Røstum et al. (1997) advised us to choose pipes that are

100m long so that the external conditions for the same pipe will be the same as well. Eisenbeis

(1999) found out that the hazard function is of a similar proportion to the square root of length.

22

Pipe material

Cast iron pipes (i.e. grey cast iron and ductile iron pipes) are used in a great number of water works

despite the fact that they have long been notorious for their high failure rates. This can also explain

the increasing use of new materials such as PVC and PE in water networks. The material features

of these pipes vary a great deal from each other, and analysis of different materials must be done

separately. Recent studies have been focusing on analyzing pipes that are made of PVC and PE in

a statistical way (Eisenbeis et al., 1999). The past few decades have witnessed great improvements

in the techniques used in the manufacturing of different pipe material. One of the best examples

showing this can be found in the improvement of the casting method used in the manufacturing of

for grey iron pipes. At the beginning, pipes were cast in sand molds in a horizontal order, which

makes the thickness of the wall become uneven. It is only after the introduction of the vertical

casting technique that the production of walls of the same thickness became possible. This new

technique has also helped to make the manufacturing of pipes with thinner walls become possible.

The improvements obtained in the centrifugal casting methods has also helped to strengthen pipes

and to help the walls to reach a higher consistency of thickness (WRc, 1998).

Seasonal variation

Winter is the season when most of the water distribution networks become the most prone to

failures. Andreou (1986) is the first person to find out that it will be easier for pipes of a smaller

diameter (those whose length do not exceed 8 inches) to break during winter. After analyzing five

23

water networks in Sweden, Sundahl (1996) found out that among the temperature of air,

precipitation and the depth of snow, only the former one would exert an effect on the break rate.

In Trondheim, even though the coldness in winter has brought forth a huge amount of frost, the

number of reported failures in summer time still overrode that for the winter season (Røstum,

1997). However, Sægrov et al. (1999) found out that the break rate in both summer time and winter

time in the United Kingdom was rather high. As the clay soils during the summer season became

increasingly drier and kept shrinking, the break rate also kept rising up, whereas during the winter

season, usually there would be a great deal of frost, and this is one of the major causes of the high

break rate. Another factor contributing to this is the thermal contraction effects. Other than this, it

is also found that the mean temperature during the day as well as the amount of rainfall each year

have also played a part in the annual break rate over a period of ten years. It is suggested that we

ought to use the effects of the climate to find out the factors leading to the failures of pipes.

However, since we do not have an idea as to how this factor change over time, it will be really

difficult for us to use the effects of the climate as a tool to forecast future failures. In her research,

Sundahl (1996) attempted to use a sinus curve to model the changes in the leakage in different

seasons. The manager held the view that the change of the failing rate of pipes according to

different seasons can offer us help to plan/organize the water network on a daily basis. However,

when it comes to the calculation of the future needs for rehabilitation and for making priorities

between pipes, the knowledge of the actual day of failure becomes less helpful.

24

Soil conditions

Soil conditions can not only exert an influence on the rate of external corrosion, but can also affect

pipe degradation. In their research, Clark et al. (1982) tried to put pipes in corrosive soil

environments and then analyzed their failing rate. They found out that how much of the pipe is

laid in corrosive environments has no relation with its breaking rate. Malandain et al. (1998) tried

using a geographic information system (GIS) to relate soil conditions to the failing rate for pipes

in the water network in Lyon, France. In his analysis of the breaking rate of pipes, Eisenbeis (1994)

used ground condition, (which is defined as the presence or absence of corrosive soil) as an

explanatory variable.

Previous failures

The braking rate of pipes in the past can help a great deal in the forecasting of future failures.

Andreou (1986) used the Cox proportional hazards model to analyze failures in the water network.

It is only after the third failure that the failing rate stopped rising, with each failure, and yet the

rate still remained to be a high one. The assumption is that at this phase, the pipes have entered a

“rapid failing state”. It is found out that failures happened in the past can exert a huge effect on the

hazard function of the pipes. Eisenbeis (1994) has also spotted a similar pattern. Malaindain et al.

(1999) has applied these findings from Andreou and Eisenbeis in a failing rate model. Goulter and

Kanzemi (1988) made close observation of the temporal and spatial gathering of water-main

breaks, which shows that it is highly likely that failures of a pipe in the past will lead to future

25

failures in its surroundings. Approximately 60% of all of the subsequent failures happened during

the first three months after the first failure. This has led us to believe that the damage brought by

the repairing work is the culprit behind these subsequent breaks. Possible damages include the rise

of pressure brought by pipe-refilling, the change of position of the ground during excavation, the

back-filling procedure or the movement of weighty vehicles. Sundahl (1996, 1997) has also

pointed out that maintenance work done on the network including repair and replacement after a

failure can also lead to a higher failing rate.

Other factors that do not share any correlation with the repair work also play a role in the

subsequent failures in the network. Pipes in the same place are usually of the same age, and the

materials that they were made from, very often, are also the same. What’s more, they are usually

constructed and jointed together via the same method. Other than all these, it is also highly possible

that both the external and internal factors that can lead to corrosion for these pipes are the same.

Nearby excavation

Excavation work done near the pipelines can exert a negative effect on the bedding conditions,

which can cause the pipe to break. Researches conducted in the U.K. (WRc, 1998) indicated that

work on closely related services (e.g. gas, electricity) can lead to pipe breaks.

The pressure in static water and the rise of pressure in a distribution system also play a role in pipe

breaks. The rise of pressure is usually caused by the opening and closing of water and air valves

while the network is under operations. These changes can be seen as one of the causes of break

26

clustering. Andreou (1986) found that when it comes to modelling pipe breaks, it can be useful to

take into account the effect of static pressure, but this factor is by no means of huge importance.

When Clark et al. (1982) were modelling time to the first break, they used both the absolute

pressure and the pressure differential (surge).

Land use

Land use (e.g. traffic zones, places of residence, and commercial areas) is used as a substitute for

external loads on pipes. Eisenbeis (1997) used land use over the pipe (i.e. no traffic vs. heavy

traffic), as a variable in break models.

Previous papers discussed a lot about the classifications and definitions of factors. However, the

availability of factors in data are limited based on real dataset. The factors have higher availability

are more likely to be considered in real models and effect more to failure prediction. The frequency

of factors using in collected dataset will be discussed in later section.

3.3 Summary of Factors Considered in Different Failure Models

27

Table 4. Considered Factors Affecting Water Pipes Failure

Classification Model References Response Age Length Diameter Installation

Year Temperature Depth

Soil

Type

Water

Press

Freezing

Index

Pipe

thickness

Previous

number

of failure

Deterministic

Models

Failure

rate

Amarjit Singh

(2012)

Failure

rate 2 1

Andreas

Scheidegger

(2017)

Average

number of

failure

2

Małgorzata

Kutyłowskaa

(2016)

Average

number of

failure

2 2

Andrew Wood

(2009)

Number

of failure 2 1

Hossein Rezaei

(2015)

Number

of failure 1 1 1 1 1

Alex

Francisque

(2017)

Failure

rate 2 2 1

Katarzyne

Pietrucbe

(2015)

Number

of failure 1 1 1 1 1 1

C.Vipulanandan

(2012)

Number

of failure 2 1

27

28

Table 4 (Cont.)



Soil

Type

Water

Press

Freezing

Index

Pipe

thickness

Previous

number

of failure

Probabilistic

Models

Weibull

proportional

hazard

model

E. Kimutai

(2015)

Number

of failure 2 1

Yves Le gat

(2000)

Failure

rate 1 1

Cox

proportional

model

H Shin

(2016)

Number


Weibull-

Based

Failure

Models

Lindsay

Jenkins

(2014)

Average

number of

failure

1 1 1

Stefano

Alvisi (2008)

Number

of failure 1 1 1

Weibull

Accelerated

lifetime

model

André

Martins

(2013)

Average

number of

failure

1 1 1 1

Weibull/Exp

onential/Exp

onential

model

Babacar

Toumbou1

(2013)

Number

of failure 2 1

Weibull/Exp

onential

model

Ben Ward

(2016)

Number

of failure 2 1

28

29

Table 4 (Cont.)



Soil

Type

Water

Press

Freezing

Index

Pipe

thickness

Previous

number

of

failure

Probabilistic

Models

Principal

component

regression

Zhiguang

Niu (2017)

Failure

rate 1 1

Multiple

regression

model

Pengjun

Yu (2013)

Number

of failure 1 1 1

Mohamed

Fahmy

(2009)

Number

of failure 1 1

Yong

Wang

(2009)

Failure

rate 1 1

Leila Dridi

(2009)

Average

number

of failure

2 2

Ahmad

Asnaashari

(2013)

Number

of failure 1 1 1 1

Kang Jing

(2012)

Number


Logistic

regression

Boxall

(2013)

Number

of failure 2 1

Non-linear

regression

B. García-

Mora

(2015)

Number

of failure 2 1

29

30

Table 4 (Cont.)



Soil

Type

Water

Press

Freezing

Index

Pipe

thickness

Previous

number

of

failure

Probabilistic

Models

Evolutionary

polynomial

regression

L. Berardi

(2008)

Number


D. A. Savic

(2009)

Number

of failure 2 1

Seyed Farzad

Karimian

(2015)

Number

of failure 1 1 1

Konstantinos

Kakoudakis

(2017)

Number


Qiang Xu

(2011)

Number

of failure 1 1 1

Fulvio Boanoa

(2015)

Number


Bayesian

method

Kleiner,Yehuda

(2012)

Number


G Kabir (2015) Number


Ángela

Martínez-

Codina (2015)

Number

of failure 2 1

30

31

Table 4 (Cont.)



Soil

Type

Water

Press

Freezing

Index

Pipe

thickness

Previous

number

of

failure

Stochastic

Models

NHPP

Peter D.

Rogers

(2009)

Number

of failure 1 1 1

T.

Economou

(2008)

Number

of failure 1 1 1

T.

Economou

(2012)

Number

of failure 1 1 1

Li Chik

(2016)

Average

number

of failure

2 2

Fengfeng

Li (2011)

Number

of failure 2 1

Yehuda

Kleiner

(2010)

Failure

rate 1 1

Poisson

process

Theodoros

Economou

(2010)

Number

of failure 1 1 2

Linear

extended

Yule

process

Yves Le

Gat (2013)

Failure

rate 1 1

Li Chik

(2016)

Average

number

of failure

2 2

31

32

Table 4 (Cont.)



Soil

Type

Water

Press

Freezing

Index

Pipe

thickness

Previous

number

of

failure

Artificial


Learningd

ANN

RaedJafar

(2010)

Number

of failure 1 1 1

M. Tabesh

(2009)

Average

number

of failure

2 2

Richard

Harvey

(2014)

Number

of failure 1 1 1

Libi P.

(2016)

Average

number

of failure

2 2

Genetic

programming

Qiang Xu

(2011)

Number

of failure 1 1 1

Wen-

zhong Shi

(2013)

Failure

rate 1 1

Fuzzy

Clustering

Mahmut

Aydogdu

(2014)

Average

number

of failure

2 2

32

33

Table 4 (Cont.)



Soil

Type

Water

Press

Freezing

Index

Pipe

thickness

Previous

number

of

failure

Artificial


Learning

Fuzzy

Clustering

Małgorzata

Kutyłowska

Average

number

of failure

2 2

Dirichlet

process

mixture of

hierarchical

beta process

model

Peng Li

(2015)

Number

of failure 1 1 1

Moran’s I

Ripley’s K-

statistic

Qiang Xu

(2012)

Number

of failure 1 1 1

Grey

relational

analysis(GRA)

Kang Jin Number

of failure 1 1 1 1

33

34

Number 2 in the form refer to use as group whereas number 1 means it is a covariate in the models.

Failure rate defined as λ which is determined from operational data using number of pipe failures

in unit time interval divide average pipeline length in a time period and the observation time.

It can be seen from Table 4 that diameter, length, and age are considered most frequently in the

network failure models. Material mostly used as cohorts or groups in the models such as NHPP,

Failure rate model, and Weibull distribution model. Diameter and length are easy to quantify and

thus are often used as covariates. These features are mostly analyzed as covariates in failure models.

Soil type is another common factor that was often included due to its availability. Although soil

type is often shown to significantly affect pipe performance, in some of the cases like Berardi

(2008), soil type is not found significant. Comparing to other factors, pipe material is usually

considered as a cohort, and different models are developed for each material cohort (add those

case). For response variable, number of failure account a large percentage.

3.4 Factor Distribution Analysis

In some cases, factors such as diameter and length are considered as covariates in the models while

material is considered as cohorts. However, sometimes, especially for failure rate models, all the

factors are considered as cohorts and the results of failure rate only apply to certain groups of pipes.

Thus, it is difficult to reach conclusion about the whole network failure status based on many

independent failure rates for different groups. The distribution of each factor is necessary to be

35

considered because the weight of each diameter and material have different weight in the modeling

results. Another important attribute for pipe is the installation year which reflects the variation in

failures over time. Since data availability of pipe failure is limited, it would be useful to know the

material installation year which has a large percentage in the whole network. The data and figures

are presented in Appendix II.

.

36

Chapter 4 Summary of Key Findings in Previous Studies

Most failure prediction models, particularly deterministic, statistical and artificial intelligence

models, characterize the relationship between network features and failures of a single network,

and predict the number of failures or life span of the network based on these factors. In this section,

we summarize the finding in collected papers and extract common conclusion about factors and

models as insights for future fitting.

4.1 Factor Effect in Regression Models

Linear models usually had a similar response variable like failure rate or break time which could

reflect the degree of failure directly. Most papers provided the result equation, so the coefficient

of parameter is easy to obtain, By this way, the results of linear model is analyzed independently

in this section. A Figure about linear model results is shown below.

37

Table 5. Results of Regression Models

References Result Equation Response

Variable

Genevieve

Pelletier (2003)

Log10R = 4.85 − 0.0206A + 0.000245A2

+ 0.00281S − 0.905Log10L

− 1.40Log10L2 − 1.40Log10S

Failure Rate

Log10R = 1.83 − 0.911Log10L

Failure Rate

Log10R = 2.69 − 0.898Log10L

− 0.745Log10A Failure Rate

Pengjun Yu

(2013)

R = 2.096 − 4.4423D + 3.3571D2

− 0.7292D3 Failure Rate

Kang Jing

(2012)

Y = −6000.741 − 1999.02D + 17318.428H

+ 450.949P2 Leakage

Time

Boxall et

al. (2013) γ(D,L,A) = 0.50247 - 0.00726D +

0.66252logL - 0.03375A + 0.00016A²

Burst Rate

In the result equation, R refers to the rate of failure; A is age of pipes; D is diameter of pipes; S is

soil type; P is water press in pipe; H is depth; L refers to length.

In the case by Pelletier (2003), the pipes of shorter lengths have higher annual break retes than

those of longer lengths. The annual break rates of the 100m length of gray cast iron pipes with

different diameters versus pipe age. In this network, 100 mm size pipe have the highset annual

break rates compared to others for all ages. The 300 and 150 mm size pipe have similar annual

break rates. For the network in case by Yu (2013), the models show a negative correlation between

38

the failure rate and diameter when the pipe diameter is less than 100mm while the failure rate is

rising when the pipe diameter is greater than 1000mm. So, the pipeline with diameter as 1000mm

has the lowest failure rate value. Jing (2012) gives a result that depth has a positive relationship to

leakage time and a negative one for diameter. The diameter limited in 50-250 mm. Boxall et al.

(2013) indicates that the burst rate only applied in certain material. Diameter, length, age is

involved in the equation.

The relationship between annual burst rate, length and diameter for cast iron and asbestos cement

pipe groups for each of the two datasets are similar, with slight variation in the coefficient values.

Once the models have been derived for a given company or region it is possible to make predictions

for every combination of material, diameter, length and age of pipe. These can be used directly to

inform investment decision making and planning, or to inform whole life cost decision support

procedures and software. It is important to recognize that this kind of burst rate prediction is valid

principally for the short term, from perhaps 1 to 5 years. The prediction for a pipe of a given age

is for its burst rate in the next year.

4.2 Model Results Review

In this section, the conclusions in the collected papers are summarized and shown in Appendix

III. Although each case has its own characteristics, the similar conclusion about model

performance or factor effect could be common. The common parts in conclusions are extracted

and shown in Table 6.

39

Table 6. Extracted Common Conclusions.

Category Common Conclusions References

Model

WE model may have a good prediction performance though it does not

consider covariates.

Toumbou et

al.(2013)

Extending WE model to WEE model or developing a WE based proportional

model would be feasible ways to improve the prediction accuracy.

Francis et al.

(2014)

However, too much covariates covered in proportional model may lead to

overfitting.

Davis et al.

(2007)

For failure rate model, modeling the failure in group and individual pipe

level would be a good way to avoid the inference that all covariates have the

same impact on pipes.

Mahmut

Aydogdu

(2014)

Cox-PHM and Poisson process both have their advantages in certain

conditions.

García-Mora et

al. (2015)

Although in most situation, Poisson process is used as a comparison for other

models or used for a small number of breaks prediction.

Asnaashari et al.

(2013)

Artificial Neural Network are useful for modeling complex problem that a

large number of covariates are included and the correlation between

covariates are uncertain.

Kabir et al.

(2015a)

Linear models usually have a lot of significant covariates and has accurate

prediction when pipe failure history is known. Otherwise, short-term

prediction would be more reliable.

Kutyłowskaa et

al. (2016)

Material

Ductile pipe has a higher failure rate when the previous number of breaks is

zero.

García-Mora et

al. (2015)

After first break, it will decrease the probability of failure, especially for

ductile pipe with long length and small diameter.

Rezaei et

al.(2015)

PVC and AC pipes suffered more from cracking which may relate to

covariates such as internal pressure, soil deflection and residual pressure.

Kleiner and

Rajani (2008)

Steel and grey cast iron suffered material corrosion which may relate to

temperature and humidity

Wood and Lence

(2009)

Time-linear model fits better than time-exponential model for as asbestos

cement (AC) and ductile iron (DI), PVC pipes usually has small number of

failures and lack of recorded history because of near installation year. So,

they can be good predicted by Poisson process.

Aydogdu and

Firat (2014)

Diameter

Diameter is a common and efficient group for failure prediction. Kimutai (2015)

Smaller diameter (25-50mm) pipe more likely to get damage which may due

to pressure fluctuation.

Kleiner and

Rajani (2008)

For pipe has high brittleness, like AC or PVC, failure rate is higher in winter

than summer, but this covariate has strong correlation with pipe-laying depth

which effect the temperature of pipes.

Martins et

al.(2013)

Generally, high pressure variation will increase failure rate. M. Tabesh

(2009)

Age, diameter, length, material, buried depth and elevation of pipe were

selected as the most critical factors.

Jenkins et

al.(2014)

Pipe diameter and age are the most sensitive factors in two datasets. Martins et

al.(2013)

For linear, NOPNF has more important weight than in other models. Karimian et

al.(2015)

Installation site has relationship with many factors. Jenkins et

al.(2014)

The relationship between burst rate and diameter has been found to increase

exponentially with decreasing diameters.

Achim et al.

(2007)

40

Chapter 5 Summary and Conclusion

Unlike previous model review papers that mostly reviewed the model development or

improvement in a period of time, this paper focuses on review and summary of contributing factors

considered in the models and the associated effects of these factors. Specifically, the characteristics

of all the collected cases are summarized to find the distribution and tendency of available

networks data; the results of different fitting models are summarized to find common conclusions.

5.1 Conclusions

Based on the review, we reach the following conclusions.

• The distribution of case regions has shown that the United States has concentrated much on

network deterioration issue. Prior to Year 2000, the case from Canada and Europe accounted

for a majority of total number of cases in papers about failure models because Canada faced

the failure problem earlier. However, the increasing number of cases in Asia and North

America indicates that some other areas started facing and solving this global common issue.

Recent papers applying to the cases in the U.S. would offer more references to future than

those applying Canadian cases.

• The analysis has also shown that the number of pipe breaks and the number of pipe segments

do not have high correlation. Thus, judging the severity of pipe deterioration based on the

number of breaks is not a feasible way.

•

41

• The summary statistics of models used in the literature has shown the popularity of prediction

models. Data driven methods has been used increasingly.

The following tables summarize the recommendations for model selection or validation of results

for future studies.

Table 7. Preferred Record Period for Material

Material Preferred Record Period

CI All

PVC After 1950

AC 1950-1970

Steel After 2000

DI 1900-2000

Table 8. Related Conclusions for Covariates.

Covariates Related Conclusions

Material Mostly used as cohorts and the number of type is not necessary to be

much

Age Has negative correlation to failure.

Diameter

Diameter less than 250mm has a negative correlation with failure,

while diameter of 1000 or above has a positive correlation with

failure.

Length Has a negative correlation with failure.

Buried depth Has a positive correlation with failure.

Pipe inner pressure Not enough conclusion.

Table 9. Preferred Condition for Models.

Models Preferred Condition

WEE Small number of covariates

Failure rate model When the input and output are clear

ANN With a large number of covariates.

Linear model When pipe failure history is known; Short term prediction.

Cox-PHM and

Poisson process

Use as comparison for models with covariate or non-covariates,

respectively.

42

5.2 Limitation and Future Work

The sample size of collected cases is limited which leads to a limitation for the analysis correlations

between model and network characteristics. Less than 20 cases offer the data of factor distributions

and the summarized conclusions may have unknown application range, e.g. the model fitting may

get influenced by network size. Thus, the conclusions of this paper are not accurate enough to be

used as verification for future model fitting.

In addition, this paper only discussed the case information, characteristics distribution and model

results separately. The link among them are not explored due to lack of time and data. So it would

be a feasible direction to do more research on the characteristics identification in network failures.

43

References

Aydogdu, M., & Firat, M. (2015). Estimation of failure rate in water distribution network using

fuzzy clustering and LS-SVM methods. Water resources management, 29(5), 1575-1590.

Alvisi, S., & Franchini, M. (2010). Comparative analysis of two probabilistic pipe breakage

models applied to a real water distribution system. Civil Engineering and Environmental

Systems, 27(1), 1-22.

Asnaashari, A., McBean, E. A., Gharabaghi, B., & Tutt, D. (2013). Forecasting watermain failure

using artificial neural network modelling. Canadian Water Resources Journal, 38(1), 24-33.

AWWA. (2001) Dawn of the Replacement Era: Reinvesting in Drinking Water Infrastructure,

AWWA Government Affairs Office, Washington, DC

Berardi, L., Giustolisi, O., Kapelan, Z., & Savic, D. A. (2008). Development of pipe deterioration

models for water distribution systems using EPR. Journal of Hydroinformatics, 10(2), 113-126.

Boano, F., Scibetta, M., Ridolfi, L., & Giustolisi, O. (2015). Water distribution system modeling

and optimization: a case study. Procedia Engineering, 119, 719-724.

Boxall, J. B., O'Hagan, A., Pooladsaz, S., Saul, A. J., & Unwin, D. M. (2007, June). Estimation of

burst rates in water distribution mains. In Proceedings of the Institution of Civil Engineers-Water

Management (Vol. 160, No. 2, pp. 73-82). London:[Published for the Institution of Civil Engineers

by Thomas Telford Ltd.], c2004-.

Chik, L., Albrecht, D., & Kodikara, J. (2016). Estimation of the Short-Term Probability of Failure

in Water Mains. Journal of Water Resources Planning and Management, 143(2), 0401607

Dey, P. K. (2003). Analytic hierarchy process analyzes risk of operating cross-country petroleum

pipelines in India. Natural Hazards Review, 4(4), 213-221. Data driven water pipe failure

prediction: A bayesian nonparametric approach. In Proceedings of the 24th ACM International on

Conference on Information and Knowledge Management (pp. 193-202). ACM.

Dridi, L., Mailhot, A., Parizeau, M., & Villeneuve, J. P. (2009). Multiobjective approach for pipe

replacement based on bayesian inference of break model parameters. Journal of Water Resources

Planning and Management, 135(5), 344-354.

Economou, T., Kapelan, Z., & Bailey, T. (2008). A zero-inflated Bayesian model for the prediction

of water pipe bursts. In Water Distribution Systems Analysis 2008 (pp. 1-11).

44

Economou, T. (2010). Bayesian modelling of recurrent pipe failures in urban water systems using

non-homogeneous Poisson processes with latent structure.

Economou, T., Kapelan, Z., & Bailey, T. C. (2012). On the prediction of underground water pipe

failures: zero inflation and pipe-specific effects. Journal of Hydroinformatics, 14(4), 872-883.

Fahmy, M., & Moselhi, O. (2009). Forecasting the remaining useful life of cast iron water

mains. Journal of performance of constructed facilities, 23(4), 269-275.

Francis, R. A., Guikema, S. D., & Henneman, L. (2014). Bayesian belief networks for predicting

drinking water distribution system pipe breaks. Reliability Engineering & System Safety, 130, 1-

11.

Francisque, A., Tesfamariam, S., Kabir, G., Haider, H., Reeder, A., & Sadiq, R. (2017). Water

mains renewal planning framework for small to medium sized water utilities: a life cycle cost

analysis approach. Urban Water Journal, 14(5), 493-501.

García-Mora, B., Debón, A., Santamaría, C., & Carrión, A. (2015). Modelling the failure risk for

water supply networks with interval-censored data. Reliability Engineering & System Safety, 144,

311-318.

Harvey, R., McBean, E. A., & Gharabaghi, B. (2013). Predicting the timing of water main failure

using artificial neural networks. Journal of Water Resources Planning and Management, 140(4),

425-434.

Jafar, R., Shahrour, I., & Juran, I. (2010). Application of Artificial Neural Networks (ANN) to

model the failure of urban water mains. Mathematical and Computer Modelling, 51(9), 1170-1180.

Jenkins, L., Gokhale, S., & McDonald, M. (2014). Comparison of Pipeline Failure Prediction

Models for Water Distribution Networks with Uncertain and Limited Data. Journal of Pipeline

Systems Engineering and Practice, 6(2), 04014012.

Jing, K., & Zhi-Hong, Z. (2012). Time prediction model for pipeline leakage based on grey

relational analysis. Physics Procedia, 25, 2019-2024.

Kabir, G., Tesfamariam, S., Loeppky, J., & Sadiq, R. (2015). Integrating Bayesian Linear

Regression with Ordered Weighted Averaging: Uncertainty Analysis for Predicting Water Main

Failures. ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part A: Civil

Engineering, 1(3), 04015007.

Kakoudakis, K., Behzadian, K., Farmani, R., & Butler, D. (2017). Pipeline failure prediction in

water distribution networks using evolutionary polynomial regression combined with K-means

45

clustering. Urban Water Journal, 14(7), 737-742.

Karimian, S. F. (2015). Failure Rate Prediction Models of Water Distribution Networks (Doctoral

dissertation, Concordia University).

Kleiner, Y., & Rajani, B. (2001). Comprehensive review of structural deterioration of water mains:

statistical models. Urban water, 3(3), 131-150.Le Gat, Y. (2014). Extending the Yule process to

model recurrent pipe failures in water supply networks. Urban Water Journal, 11(8), 617-630.

Kleiner, Y., & Rajani, B. (2010). I-WARP: Individual water main renewal planner. Drinking

Water Engineering and Science, 3(1), 71.

Kleiner, Y., & Rajani, B. (2012). Comparison of four models to rank failure likelihood of

individual pipes. Journal of Hydroinformatics, 14(3), 659-681.

Kimutai, E., Betrie, G., Brander, R., Sadiq, R., & Tesfamariam, S. (2015). Comparison of

statistical models for predicting pipe failures: Illustrative example with the City of Calgary water

main failure. Journal of Pipeline Systems Engineering and Practice, 6(4), 04015005.

Kutyłowska, M., & Orłowska-Szostak, M. (2016). Comparative analysis of water–pipe network

deterioration–case study. Water Practice and Technology, 11(1), 148-156.

Kutylowska, M. (2015). Modelling of failure rate of water-pipe networks. Periodica Polytechnica.

Civil Engineering, 59(1), 37.

Le Gat, Y., & Eisenbeis, P. (2000). Using maintenance records to forecast failures in water

networks. Urban Water, 2(3), 173-181.

Li, F., Ma, L., Sun, Y., & Mathew, J. (2014). Group maintenance scheduling: a case study for a

pipeline network. In Engineering Asset Management 2011 (pp. 163-177). Springer, London.

Lin, P., Zhang, B., Wang, Y., Li, Z., Li, B., Wang, Y., & Chen, F. (2015, October).

Martins, A., Leitão, J. P., & Amado, C. (2013). Comparative study of three stochastic models for

prediction of pipe failures in water supply systems. Journal of Infrastructure Systems, 19(4), 442-

450.

Markose, L. P., & Deka, P. C. (2016). ANN and ANFIS Modeling of Failure Trend Analysis in

Urban Water Distribution Network. In Urban Hydrology, Watershed Management and Socio-

Economic Aspects (pp. 255-264). Springer International Publishing.

Martínez-Codina, Á., Cueto-Felgueroso, L., Castillo, M., & Garrote, L. (2015). Use of pressure

46

management to reduce the probability of pipe breaks: A Bayesian approach. Journal of Water

Resources Planning and Management, 141(9), 04015010.

Mehrani, M. J., Jalili, M. R., Abadi, Y. T., Tashayoei, M. R., Hanifi, H., & Yusefpour, R. New

equations for Prediction of pipe burst rate in water distribution networks.

National Research Council. (2006). Drinking water distribution systems: assessing and reducing

risks. National Academies Press.

Niu, Z., Wang, C., Zhang, Y., Wei, X., & Gao, X. (2017). Leakage Rate Model of Urban Water

Supply Networks Using Principal Component Regression Analysis. Transactions of Tianjin

University, 1-10.

Pelletier, G., Mailhot, A., & Villeneuve, J. P. (2003). Modeling water pipe breaks—Three case

studies. Journal of water resources planning and management, 129(2), 115-123Rajeev, P.,

Kodikara, J., Robert, D., Zeman, P., & Rajani, B. (2014). Factors contributing to large diameter

water pipe failure. Water asset management international, 10(3), 9-14.

Pietrucha-Urbanik, K. (2015). Failure analysis and assessment on the exemplary water supply

network. Engineering Failure Analysis, 57, 137-142.Vipulanandan, C., Qiao, W., & Hovsepian, H.

(2012). Water Pipeline Failures in the Active Zone. Center for Innovative Grouting Materials and

Technology (CIGMAT) 2012.

Rajani, B., & Kleiner, Y. (2001). Comprehensive review of structural deterioration of water mains:

physically based models. Urban water, 3(3), 151-164.

Rezaei, H., Ryan, B., & Stoianov, I. (2015). Pipe failure analysis and impact of dynamic hydraulic

conditions in water supply networks. Procedia Engineering, 119, 253-262.

Rogers, P. D., & Grigg, N. S. (2009). Failure assessment modeling to prioritize water pipe renewal:

two case studies. Journal of Infrastructure Systems, 15(3), 162-171.

Sargaonkar, A., Kamble, S., & Rao, R. (2013). Model study for rehabilitation planning of water

supply network. Computers, Environment and Urban Systems, 39, 172-181.

Savic, D. A., Giustolisi, O., & Laucelli, D. (2009). Asset deterioration analysis using multi-utility

data and multi-objective data mining. Journal of Hydroinformatics, 11(3-4), 211-224.

Singh, A., & Adachi, S. (2011). Expectation analysis of the probability of failure for water supply

pipes. Journal of Pipeline Systems Engineering and Practice, 3(2), 36-46.

Shin, H., Kobayashi, K., Koo, J., & Do, M. (2016). Estimating burst probability of water pipelines

47

with a competing hazard model. Journal of Hydroinformatics, 18(1), 126-135.

Shi, W. Z., Zhang, A. S., & Ho, O. K. (2013). Spatial analysis of water mains failure clusters and

factors: a Hong Kong case study. Annals of GIS, 19(2), 89-97.

Tabesh, M., Soltani, J., Farmani, R., & Savic, D. (2009). Assessing pipe failure rate and

mechanical reliability of water distribution networks using data-driven modeling. Journal of

Hydroinformatics, 11(1), 1-17.

Toumbou, B., Villeneuve, J. P., Beardsell, G., & Duchesne, S. (2012). General model for water-

distribution pipe breaks: Development, methodology, and application to a small city in Quebec,

Canada. Journal of Pipeline Systems Engineering and Practice, 5(1), 04013006.

Scheidegger, A., Scholten, L., Maurer, M., & Reichert, P. (2013). Extension of pipe failure models

to consider the absence of data from replaced pipes. Water research, 47(11), 3696-3705.

Ward, B., Selby, A., Gee, S., & Savic, D. (2017). Deterioration modelling of small-diameter water

pipes under limited data availability. Urban Water Journal, 14(7), 743-749.

Wang, Y., Zayed, T., & Moselhi, O. (2009). Prediction models for annual break rates of water

mains. Journal of performance of constructed facilities, 23(1), 47-54.

Wood, A., & Lence, B. J. (2009). Using water main break data to improve asset management for

small and medium utilities: District of Maple Ridge, BC. Journal of Infrastructure Systems, 15(2),

111-119.

Xu, Q., Chen, Q., Li, W., & Ma, J. (2011). Pipe break prediction based on evolutionary data-driven

methods with brief recorded data. Reliability Engineering & System Safety, 96(8), 942-948.

Xu, Q., Chen, Q., & Li, W. (2011). Application of genetic programming to modeling pipe failures

in water distribution systems. Journal of Hydroinformatics, 13(3), 419-428.

Yu, P. J., Nie, J. X., Xu, G., Long, Z. H., Wang, Z. H., Tu, Y., ... & Chen, Y. L. (2013). Study of

Failure Rate Model for a Large-Scale Water Supply Network in Southern China Based on

Different Diameters. In Applied Mechanics and Materials (Vol. 260, pp. 1200-1205). Trans Tech

Publications.

48

Appendix I

Table 10. Information Summary of Collected Cases

# Title References Area Country City Population

Network

length (km)

Number of

pipe

segment

Number of

pipe

breaks

1

Modeling Water Pipe Breaks—Three

Case Studies

Genevieve

Pelletier (2003)

North

America Canada Chicoutimi 64000 352 2096 1719

2

North

America Canada Gatineau 93000 407 1554 1426

3

North

America Canada Saint-Georges 20000 155 1806 279

4

Failure Assessment Modeling to

Prioritize Water Pipe Renewal: Two

Case Studie

Peter D. Rogers

(2009)

North

America America Colorado Spring 400000 2900 1471 1771

5

North

America America Laramie Water 30000 330 3792 667

6

Development of pipe deterioration

models for water distribution systems

using EPR

L. Berardi

(2008) Europe UK

48 water quality

zones 19494 173 3669 354

7

Application of Artificial Neural

Networks (ANN) to model the failure

of urban water mains

RaedJafar

(2010) Europe France 43000 162 4862

8

A zero-inflated Bayesian models for

the prediciton of water pipe bursts

T. Economou

(2008)

North

America Canada

South-Central

Ontario 1349 5425

9

On the prediction of underground

water pipe failures: zero inflation and

pipe-specific effects

T. Economou

(2012) Asia New Zealand 532 175

10

Integrating Bayesian Linear

Regression with Ordered Weighted

Averaging: Uncertainty Analysis for

Predicting Water Main Failures G Kabir (2015)

North

America Canada Calgary 1100000 4281km 49531

11

Comparative Study of Three

Stochastic Models for

Prediction of Pipe Failures in Water

Supply Systems

André Martins

(2013) Europe Portugal 367km 11472 1912

48

49

Table 10 (Cont.)


Network

length

(km)

Number of

pipe segment

Number

of pipe

breaks

12

Expectation Analysis of the Probability of

Failure for Water Supply Pipes

Amarjit Singh

(2012) North America US

Island of

Oahu 3200

13

General Model for Water-Distribution Pipe

Breaks:

Babacar

Toumbou

(2013) North America Canada

City in

Quebec, 185km 1152

14

Comparison of Statistical Models for Predicting

Pipe Failures: Illustrative Example with the City

of Calgary Water Main Failure

E. Kimutai

(2015) North America Canada Calgary 149552

15

Estimation of the Short-Term Probability

of Failure in Water Mains Li Chik (2016) Asia Australia Melbourne 376

16

Assessing pipe failure rate and mechanical

reliability of

water distribution networks using data-driven

modeling

M. Tabesh

(2009) Europe Iran 93719 579

17

I-WARP: Individual water mAin renewal

planner

Yehuda Kleiner

(2010) North America

Western

Canada 146.6 1091

18

Extending the Yule process to model recurrent

pipe failures in water supply networks

Yves Le Gat

(2013) North America US Mid-Atlantic 627.2 10581 10286

19

GROUP MAINTENANCE SCHEDULING: A

CASE STUDY FOR A PIPELINE NETWORK

Fengfeng Li

(2011) Asia Australia 50000 3640 66405

20

Data Driven Water Pipe Failure Prediction: A

Bayesian

Nonparametric Approach Peng Li (2015) Asia China A 210000

23

Bayesian Belief Networks for Predicting

Drinking Water Distribution

Royce Fransis

(2014) North America US Mid-Atlantic 500000 403.4 2598 3686

24

Extension of pipe failure models to consider the

absence

of data from replaced pipes

Andreas

Scheidegger

(2017) Europe Swizerland Lausanne

49

50

Table 10 (Cont.)


Network

length (km)

Number of

pipe

segment

Number of

pipe breaks

25

Prediction Models for Annual Break Rates

of Water Mains Yong Wang (2009)

North

America Canada

Monton,

Laval, Quebec 432km

26

Estimation of Failure Rate in Water

Distribution Network

Using Fuzzy Clustering and LS-SVM

Methods

Mahmut Aydogdu

(2014)

North

America Malatya 550000 440km 1231

27

Comparative analysis of water–pipe

network deterioration–case study

Małgorzata

Kutyłowskaa(2016) Europe Poland A 40000 10.7 269

28

Estimating burst probability of water

pipelines with a

competing hazard model H Shin (2016) Asia

South

Korea 848.1km 26577 1405

29

Study of Failure Rate Model for a Large-

scale Water Supply Network in

Southern China Based on Different

Diameters Pengjun Yu (2013) Asia China

Southern

China

30

Forecasting watermain failure using

artificial

neural network modelling

Ahmad Asnaashari

(2013)

North

America Canada Toronto 784km 5422

31

Comparative analysis of two probabilistic

pipe breakage models

applied to a real water distribution system

Stefano Alvisi

(2008)

North

America Italy Ferrara 250000 2400km 23000 3472

32

Asset deterioration analysis using multi-

utility data and

multi-objective data mining D. A. Savic (2009) Europe UK 189 477 89

33

Application of genetic programming to

modeling pipe

failures in water distribution systems Qiang Xu (2011) Asia China Beijing 3322.5km 313804 566

34

Spatial analysis of water mains failure

clusters and factors: A Hong Kong case

study

Wen-zhong Shi

(2013) Asia China Hong Kong 643km 84127

35

Modelling of Failure Rate of Water-pipe

Networks

Małgorzata

Kutyłowska (2012) Europe Poland A 16km

50

51

Table 10 (Cont.)


Network

length

(km)

Number of

pipe

segment

Number of

pipe breaks

36

Using Water Main Break Data to Improve Asset

Management

for Small and Medium Utilities: District of

Maple Ridge, B.C.

Andrew Wood

(2009)

North

America Canada Maple Ridge 6000 43.55km 54

37

Water distribution system modeling and

optimization: a case study

Fulvio Boanoa

(2015) 50000 170km

38

Time Prediction Model for Pipeline Leakage

Based on Grey

Relational Analysis

Kang Jing

(2012) Asia China North China

40

Forecasting the Remaining Useful Life of Cast

Iron

Water Mains

Mohamed

Fahmy (2009)

North

America

Canada,

USA 150000 221

41

Multiobjective Approach for Pipe Replacement

Based

on Bayesian Inference of Break Model

Parameters

Leila Dridi

(2009)

42

Predicting the Timing of Water Main Failure

Using Artificial Neural Networks

Richard Harvey

(2014)

North

America Canada

Greater

Toronto

Area 5500000 5850km 6346 9918

43

Comparison of Pipeline Failure Prediction

Models

for Water Distribution Networks with Uncertain

and Limited Data

Lindsay Jenkins

(2014)

North

America USA southeastern 600000 4800km

44

Leakage Rate Model of Urban Water Supply

Networks Using Principal Component

Regression Analysis

Zhiguang Niu

(2017) Asia China Tianjin 15.17 mi 5000km

45

Deterioration modelling of small-diameter water

pipes under limited data availability

Ben Ward

(2016) Europe 1.5 mi 15000 800000 60827

46

ANN and ANFIS Modeling of Failure

Trend Analysis in Urban Water

Distribution NetworkANN and ANFIS

Modeling of Failure

Trend Analysis in Urban Water

Distribution Network

Markose and

Deka (2016) Asia Indian Trivandrum 137714

47

Time Prediction Model for Pipeline Leakage

Based on Grey Relational Analysis*

Kang Jing

(2012)

51

52

Table 10 (Cont.)


Network

length (km)

Number of

pipe segment

Number

of pipe

breaks

48

Model study for rehabilitation

planning of water supply network

Aabha Sargaonkar

(2012)

49

Using maintenance records to

forecast failures in water networks Yves Le gat (2000) Europe France

Charente-

Maritime 1243km 1212 735

50

Pipeline failure prediction in water

distribution networks using

evolutionary polynomial regression

combined with K-means clustering

Konstantinos

Kakoudakis (2017) Europe UK

53

Estimation of burst rates in water

distribution mains Boxall (2013) Europe UK 36000 4335

54

Failure Rate Prediction Models of

Water Distribution Networks

Seyed Farzad

Karimian (2015) Asia Qatar Montreal 1.8mi 5045km 125828 22735

55

New equations for Prediction of

pipe burst rate in water distribution

networks

Mohammad Javad

Mehrani (2015) Asia Tehran

56

Comparison of four models to rank

failure likelihood

of individual pipes

Kleiner, Yehuda

(2012)

North

America A(CI) 1091

57

Pipe failure analysis and impact of

dynamic hydraulic conditions in

water supply networks

Hossein Rezaei

(2015) Europe UK 100000 1090 5427

58

Modelling the failure risk for water

supply networks with interval-

censored data

B. García-Mora

(2015) Europe Spain Mediterranean 25026 1487

52

53

Appendix II

Table 11. Diameter Percentage in Cases.

1982-

2003

1982-

2003

1982-

2003

1993-

2005

1993-

2003

2000-2006

0-100 2 3 4 5 2.8 50

100-200 82 66 65 30 57 28

200-300 10 25 16 25 19 13

300-400 5 5 15 16 22.4 7

400-600 0 0 0 23 0 2

Figure 2. Distribution of Diameters for Each Available Case

0

10

20

30

40

50

60

70

80

90

0-100 100-200 200-300 300-400 400-600

%Diameter Percentage in Cases

1 2 3 4 5 6Case:

Diameter (mm)

54

Table 12. Installed Pipes Percentage in Cases

1

1950-

2013

1950-

2013

1982-

2003

1982-

2003

1982-

2003

2000-

2006

<1945 0 0 3 0 0 3

1945-1960 46 0 25 20 10 13

1961-1975 9 18 30 45 30 12

1976-1996 19 24 35 35 60 34

1996-2010 26 56 0 0 0 38.5

Figure 3. Installed Pipes Percentage in Cases

0

10

20

30

40

50

60

70

<1945 1945-1960 1961-1975 1976-1996 1996-2010

%Installed Pipes Percentage in Cases

1 2 3 4 5 6

Year of Installation

Case:

55

Table 13. Material Percentage in Cases.

1940-

2010

1972-

2015

1992-

2003

1992-

2003

1993-

2003

1995-

2005

1999-

2012

1999-

2012

2000-

2006

2000-

2006

2003-

2013

2006-

2012

CI 20 56.5 35 44 20 15 56.6 64.1 2.9 15 69 55

DI 23 26.6 42 35 25 40 0 2 0 0 5 0

PVC 57 5.5 17 25 54 3.2 8.4 7.4 0 17 7 27.4

AC 0 10.5 0 0 0 0 1.8 0.6 31 55 10 1.8

PE 0 0 0 0 0 0 17.2 24 34 0 3 0

Steel 0 0 0 0 0 1.9 8 16 0 15.6 0 0

Figure 4. Material Percentage in Cases

0

10

20

30

40

50

60

70

80

1 2 3 4 5 6 7 8 9 10 11 12

%

Case Number

Material Percentage in Cases

CI DI PVC AC PE Steel

56

Appendix III

Table 13. Model Results Review.

References Model Main Conclusions Type

Scheidegger

et al.(2017) WE

It is obvious that the failure rate of the first-generation

ductile pipes is higher Material

Toumbou et

al.(2013) WE

The WEE mode is not affected by the covariates Model

Comparison

The effect of pipe diameter grouping is more useful in long

term failure prediction Diameter

Davis et al.

(2007) WE

For pipes made of PVC, the time to brittle fracture for pipes

with internal defects are caused by internal pressure, soil

deflection and residual stress

Material

Francis et al.

(2014) Data driven

Population density cannot be used to find the relation

between pipe age and intensity of water due to its lack of

accuracy

Population

García-Mora

et al. (2015) Data driven

Long and small pipes made of ductile cast material will not

break easily when they are put under sidewalks

Length,

Diameter,

Material

Asnaashari et

al. (2013) Failure rate

Both the CP and the CML programs can help to decrease the

failure rate

Internal

protection

Kutyłowskaa

et al. (2016) Failure rate

Change in the pressure might be one of the causes of the

damage of small pipes (25-50mm) Diameter

Grey cast iron can be influenced by corrosion; Pipes made

of AC or plastic will only be affected by cracks; Steel is

exempt from the harm of material corrosion.

Material

Pipes that are not laid deep into the ground are more likely

to break in winter time than in summer time Temperature

Rezaei et

al.(2015) Failure rate Change in pressure can lead to failure of the pipe Pressure

Kleiner and

Rajani (2008) Failure rate

Covariates at both group and pipe levels are analyzed so that

the inference that all covariates will exert the same influence

on pipes can be avoided

Model

Comparison

Wood and

Lence (2009) Failure rate

The time-linear models can help to make the results of the

analysis of pipe material groups become more accurate Material

Aydogdu and

Firat (2014) Failure rate

Pipes with a diameter of 110cm, pipes that are 0-200m long,

and pipes aging from 15 years to 20 years are the easiest to

break

Length,

Diameter,

Age

Kimutai

(2015)

Weibull

Proportioanl

Hazard

Model; Cox

Proportional

Hazard

Model

Thanks to its accurate estimation of the number of failures,

compared with Cox-PHM and Poisson process, WPHM does

a much better job at predicting the failing rate of metallic

pipes

Model

Comparison

As the failing speed of the pipes becomes increasingly

faster, the forecasting made via Cox-PHM becomes less and

less accurate

Model

Comparison

Cox-PHM is a better option for the forecasting of young

systems.

PM is a better option for the forecasting of PVC pipes Material

57

Table 13 (Cont.)


Jenkins et

al.(2014)

Weibull

Proportioanl

Hazard Model;

Cox

Proportional

Hazard Model

If we reduce the number of explanatory variables,

then it will become less likely for us to overfill a

model

Size

It can be difficult for us to learn more about the

uncertain length of segment pipe from known data Length

Karimian

et al.(2015)

Evolutionary

Polynomial

Regression

(EPR)

Length, diameter, age, material,elevation and the

buried depth of pipes were chosen as the most

important factors

Length,

Diameter,

Material

Among all the factors, age and the diameter of pipes

are the most sensitive ones in two of the data sets

Diameter,

Age

Achim et

al. (2007)

Artificial

Neural Network

(ANN)

ANNs can help a great deal in the modeling of

sophisticated problems and these models can deal

with all the effects brought by a wide range of input

variables

Model

Comparison

Both time-dependent and other supplementary

factors will be incorporated in the analysis for this

model

Model

Comparison

Kabir et al.

(2015a)

Linear

Regression

Model

CI pipes are more sensitive to the resistance of soil

while the DI pipes are more sensitive to the soil

corrosivity index

Material

Martins et

al.(2013)

linear extension

of the Yule

process

(LEYP);

Weibull

accelerated

lifetime model

(WALM)

Neither the linear-extended Yule process nor the

Weibull accelerated lifetime model can affect the

avoidable breaks

Model

Comparison

The number of past breaks are the priority for both

LEYP and WALM

Historical

Failure

Results shown by the other two models are slightly

better than the Poisson results

Model

Comparison

Without the effect of past breaks, both LEYP and

WALM would perform in as similar way as the

Poisson process does

Model

Comparison

Repair work can make a pipe become more prone to

breaks

Historical

Failure

Under the circumstances when only a small number

of variables are available, the Poisson process can

become very good at forecasting the failing rate

Model

Comparison

The shorter the maintenance records are, the better

the forecasting done by LEYP and WALM will be

Model

Comparison

Le Gat et

al.(2013) Linear

Time-dependent factors including implementation

of pipe protection measures, changes in the traffic in

the road, the time of frost as well as rainy weather

sequence can all be used as references

Other Effect

The thickness of the walls of pipes are closely

related to the break rate Thickness

58

Table 13 (Cont.)


Boxall (2013) Linear

The reduce of diameters can help to strengthen

the correlation between the burst rate and the

diameter

Diameter

Very little changes were found in the

correlation between length and the burst rate Length

The correlation between age and the burst rate

is different if we analyze it in different ways Age

Without a dependable age relationship, we

would not be able to make long-term

forecasting of burst rates.

Model

Comparison

Wang et al.

(2010) Linear

The diameter and age of a pipe are the factor

that can exert the biggest effect on the

condition of the pipe

Age,

Diameter

Since the recharge of electricity, the depth of

the trench and the number of roads share no

relation to the condition of a pipe, they were

not takne into consideration in the final

analysis

Environment

Kleiner,

Yehuda (2012) NHPP

The number of past breaks, length as well as

the age of the covariates can become very

important statistics when the NHPP based

model is in use

Age,

Length,

Historical

Failure

Compared with ductile iron pipes, cast iron

mains are more prone to the effects of factor

related to the climate

Material

C.Vipulanandan

(2012)

Genetic

Programming;EPR

Models used for big cities ought to be different

from those used for smaller cities

Network

Size

Peter D. Rogers

(2009)

multiple-

criteria decision

analysis (MCDA)

Three failures happened in the past will be

needed if we are using NHPP to do a single

forecasting of pipe break

Model

Comparison

L. Berardi

(2008)

Evolutionary

polynomial

regression

Compare with pipes with a larger diameter,

when the external pressure becomes really

strong, smaller pipes are easier to break

Diameter

T. Economou

(2008)

zero-inflated

NHPP

Compared with the general NHPP, the zero-

inflated version of the NHPP is more suitable

for the data and the results it provides is of a

slightly higher accuracy even though its

performance is still worse than the general one

Model

Comparison

T. Economou

(2012) NHPP

For pipes that remained not to break while

being watched, the use of the Zero-inflated

NHPP will give us a more accurate forecasting

Model

Comparison

G Kabir (2015) Bayesian Linear

regression

The mean response forecasting made by the

Bayesian regression models is no different

from that made by the normal regression

model, but predicted response made by the

former one is better

Model

Comparison

59

Table 13 (Cont.)


M. Tabesh

(2009)

Artificial

neural network

When it comes to the evaluation of the

mechanical reliability (availability) values, the

ANN pipe failure rate model does a much

better job than the Adaptive NeuroFuzzy

Inference System (ANFIS).

Model

Comparison

Yehuda

Kleiner

(2010)

NHPP

For pipes with nearly no failures in the past,

the aggregated total number of failures per

pipe given by the NHPP was over estimated,

while the forecasting made by the same model

for pipes have failed for many times in the

past was underestimated

Model

Comparison

Fengfeng

Li (2011)

Dirichlet

process mixture

of hierarchical

beta process

model

Pipes whose predicted likelihood of breaks in

the future do not exceed 0.1 would not be

included in future analysis

Model

Comparison

Yong

Wang

(2009)

Multiple

Regression

model

Short pipes that have broken for more times in

a year do not necessarily have more failures

than long pipes do

Length

Mahmut

Aydogdu

(2014)

Fuzzy

Clustering;

Leaset square

support vector

machine

method (LS-

SVM)

LSSVM model results for the sub-regions

defined by clustering analysis are better and

that the clustering analysis can help to

improve the performance of the estimation

model and to provide a better result

Model

Comparison

Documents

Systematic Review for Water Network Failure Models and Cases